Chapter Six - Moore’s Law Meets the Brain

CHAPTER SIX

Moore’s Law Meets the Brain

THE EXPERT SYSTEMS crashed. The logic trees rotted. The suits left the building.

But in the corners of academia, a few stubborn researchers were tinkering with something heretical. Something that had already been declared dead by the AI establishment.

Neural networks.

You know — like the brain.

They weren’t pretty. They weren’t scalable. They barely worked.

But they learned.

And as computers got faster, memory got cheaper, and data began to flood the world, these weird little digital neurons started doing something nobody expected:

Winning.

Gordon Moore, co-founder of Intel, made a prediction in 1965: the number of transistors on a chip would double every two years. He was right — for decades.

This exponential curve — known as Moore’s Law — quietly transformed everything. More transistors meant faster chips, more memory, better storage. What once took a room of machinery could now fit on a chip smaller than your fingernail.

By the mid-1990s, the hardware finally caught up with the math. Suddenly, you could run millions of calculations per second. You could simulate thousands of neurons. You could train simple models not in weeks… but in hours.

AI didn’t get smarter. The world just got fast enough to fake it.

Neural networks had been around since the 1940s. They were inspired by the brain — or at least, a cartoon version of it. You had neurons (math functions) that connected to each other, fired signals, and adjusted themselves based on feedback. That process was called backpropagation — a way for the network to learn from its mistakes.

It didn’t look like logic. It looked like intuition.

But in the 1980s, nobody could make it scale.

By the late 1990s and early 2000s? It started to work.

These weren’t your buttoned-up logic guys anymore. These were the rebels.

Geoffrey Hinton, the “Godfather of Deep Learning,” who never stopped believing in neural nets, even when the field laughed him out of the room.

Yann LeCun, who helped develop convolutional neural networks (or CNNs) for visual pattern recognition — and got them working on digits, then faces.

And Yoshua Bengio, who pushed the limits of unsupervised learning and helped lay the groundwork for generative models.

They weren’t chasing chess or language or war.

They were chasing pattern recognition — the raw ability to learn from data without needing explicit rules.

As the 2000s rolled in, something strange happened.

The internet exploded. Search engines took off. Social media erupted. Phones became sensors. Surveillance ramped up. Every click, photo, GPS ping, and emoji became data.

Suddenly, there wasn’t just enough data to train neural networks — there was too much.

And these once-ignored algorithms? They were perfectly designed to eat it.

Pattern in. Pattern out.

No explanations. No symbols. No logic.

Just results.

Still, in the early 2000s, neural networks were only beating symbolic systems on paper. They were fragile. Hard to train. Finicky. The real explosion hadn’t happened yet.

But all the ingredients were there:

• Hardware acceleration (hello, GPUs)
• Massive labeled datasets (thanks, internet)
• Backpropagation algorithms
• And researchers who refused to quit

The field was coiled tight.

It just needed a trigger.

And in 2012, it got one.